Applying zero downtime Elasticsearch index mapping changes (Ruby on Rails)

README.md

This gist assums that you are running a Ruby on Rails project and you are using the elasticsearch-rails gem to operate with your Elasticsearch cluster.

This script is using the Elasticsearch Index Aliases and Reindex API to provide a way for applying zero downtime index mapping changes (adding, deleting or modifying a field).

How it works?

Let's say that you have an Article model which by default references an articles index in Elasticsearch. And lets assume that articles is an alias to a timestamped index articles_20180101000001.

Running the script will:

1. Create a new timestamped index articles_20180101000002
2. Will use the Reindexing API to copy all documents from articles_20180101000001 to articles_20180101000002
3. Will alias articles_20180101000002 to articles.
4. Will delete articles_20180101000001

mapping.rb

model = Article
elasticsearch = Elasticsearch::Client.new host: 'http://localhost:9200',
                                          transport_options: {
                                            request: {open_timeout: 30}
                                          },
                                          logger: Rails.logger

alias_name = model.index_name
alias_exists = elasticsearch.indices.exists_alias? name: alias_name
aliased_index_names =
  if alias_exists
    elasticsearch.indices.get_alias(name: alias_name).map { |index_name, _| index_name }
  else
    []
  end
old_index_name = aliased_index_names.first
new_index_name = "#{model.__elasticsearch__.index_name}_#{Time.current.strftime('%Y%m%d%H%M%S')}"

raise 'More than one aliased indexes exist' if aliased_index_names.many?

elasticsearch.indices.create index: new_index_name,
                             body: {
                               settings: model.settings.to_hash,
                               mappings: model.mappings.to_hash,
                             }

if old_index_name.present?
  elasticsearch.reindex body: {
                          source: {index: old_index_name},
                          dest: {index: new_index_name}
                        }
else
  model.import index: new_index_name
end

if alias_exists
  elasticsearch.indices.update_aliases body: {
    actions: [
      {remove: {index: old_index_name, alias: alias_name}},
      {add: {index: new_index_name, alias: alias_name}},
    ]
  }
else
  elasticsearch.indices.put_alias index: new_index_name, name: alias_name
end

elasticsearch.indices.delete index: old_index_name